This invention relates to systems for cooling electrical components; and more particularly, it relates to systems for cooling integrated circuit packages with a liquid in a digital computer.
Initially in the prior art, integrated circuits in digital computers were cooled by convection with air. Typically, the integrated circuits of the computer were mounted on several printed circuit boards which in turn were mounted in a frame. Then one or more fans were provided within the frame to simply blow air across the integrated circuits. Such a cooling system is relatively inexpensive; however, it also has several major limitations.
For example, as the amount of circuitry on an integrated circuit chip increases, the amount of power and the amount of heat which that chip dissipates also increases. Thus a point is eventually reached with very large scale integrated circuits or with multichip integrated circuit packages at which the power dissipation is simply too high to allow cooling by air convection. Also, the logic gates in integrated circuit packages operate at slower speeds as their operating temperature is raised. Further, integrated circuits are more prone to failure when they are operated at higher temperature.
Accordingly, in the prior art, systems for cooling integrated circuits by conduction with a liquid have been developed. One such system which IBM uses for example in their 3081 computers, consists essentially of a plurality of 3081 multichip cooling modules. Each module includes a base plate, a substrate with several integrated circuit chips, a piston holding plate, and a cold plate with water channels. These items are bolted together one on top of the other in the above recited order.
Formed in the piston holding plate are several cylinders, each of which contains a helical spring and a piston. In operation, each spring pushes a piston against a respective integrated circuit chip on the substrate; and heat from each chip then travels in a serial fashion through the piston, through the cylinder sidewalls, and into the cold plate to the water.
However, cooling in this IBM module is still substantially limited because the water does not flow directly over the surface of the integrated circuit chips, and because thermal conduction between a piston and a cylinder sidewall is poor. In addition, liquid can leak from the cold plate at its input port or its output port when a defective connection is there made since the liquid passes through the cold plate under high pressure.
In another liquid cooling system, which is used on Cray-2 supercomputers, multichip circuit modules are completely immersed in a liquid bath. But this makes it cumbersome to remove a module for repair. Also, only inert liquids can be used; otherwise conductive traces which interconnect the circuit chips will corrode. Further, the liquid must have a very low dielectric constant so that electrical signals on the conductive traces do not propagate slowly. To meet these requirements, Cray-2 uses a special liquid called FC-77. But its surface tension is four times smaller than the surface tension of water; and this places considerable demands on the seals and gaskets in the cooling system--otherwise they will leak.
Accordingly, a primary object of the invention is to provide an improved liquid cooling system for integrated circuit packages in digital computers in which all of the above problems, and others, are overcome.